Prejudice and discrimination have unfavorable effects. People who frequently experience these intergroup processes have greater psychological suffering, depression, health risk behaviors, and less satisfaction with life (Gibbons et al., 2014). In addition, they have less access to resources such as health care, employment opportunities, housing, or education (Stroebe et al., 2010).

In Europe, Roma people are the largest ethnic minority group (estimated to 10–12 million people, Bernát & Messing, 2016) and are frequent targets of hate crimes and discrimination in interpersonal and institutional contexts (European Union Agency for Fundamental Rights, 2016). However, the psychosocial literature on prejudice toward Roma is scarce (for some exceptions see Kende et al., 2020; Navas & Cuadrado, 2012; Urbiola et al., 2018).

Although in many countries the situation of this group has improved, the gap between Roma and non-Roma populations still exists: Roma are systematically disadvantaged in housing, employment, education, health care, and life expectancy (Bojadjijeva, 2015); they are also the target of negative stereotypes and emotions across countries (Kende et al., 2020). This is also the situation in Spain, where Roma is the main derogated group above other ethnocultural minority groups, such as immigrants (e.g., Navas & Cuadrado, 2012; for a socio-political contextualization in Spain see Magazzini & Piemontese, 2016).

This paper will examine to what extent prejudice and discrimination toward Roma in Spain are based on ethnic, social class categories or on an interaction between these variables (see Moore-Berg & Karpinski, 2019; Weeks & Lupfer, 2004 for related research questions about other minorities).

Prejudice: Specific Stereotypes, Emotions, and Behaviors

Recent models on prejudice, such as the stereotype content model (SCM; Fiske et al., 2002) and the BIAS Map (Cuddy et al., 2008) emphasize that prejudice is a construct composed by multiple factors and has specific and different characteristic for each target group. Prejudice includes evaluative responses (both subjectively negative and positive). These responses can be cognitive (stereotypes), affective (emotions), and behavioral (behavioral tendencies) and will vary depending on specific groups and contexts (e.g., Cuddy et al., 2007, 2008; Fiske et al., 2002).

SCM (Fiske et al., 2002) considers two stereotypical dimensions that dominate our social perception of others: warmth (e.g., kind, friendly) and competence (e.g., efficient, intelligent). According to this model, Roma are perceived to be low in both warmth and competence (Grigoryev et al., 2019), and they will elicit mainly negative emotions (Fiske et al., 2002). In line with more recent research (Brambilla et al., 2012; López-Rodríguez & Zagefka, 2015), in the present work, a stereotypical three-dimensional model—morality (e.g., honesty), sociability (e.g., kindness), and competence (e.g., efficiency)—will be used because of its greater complexity and better predictive validity (Brambilla & Leach, 2014).

On the other hand, the BIAS Map (Cuddy et al., 2008) provides a general structure about how stereotypes and emotions experienced toward other groups are related to behaviors toward them. These behaviors could include positive and negative actions called in the model as behavioral tendencies of facilitation (e.g., to help, assist, collaborate), and behavioral tendencies of harm (e.g., to harm, threaten, ignore), respectively. According to BIAS Map, it is to be expected that the majority will manifest a less willingness to carry out behavioral tendencies of facilitation than of harm toward Roma.

Thus, understanding the specificity of stereotypical attributions, emotions, and behavioral tendencies, based on different group membership categories (e.g., gender, ethnic group or, in this study, social class), or the interaction between them, is key for developing effective strategies to reduce inequality and achieve more positive intergroup attitudes (Velasco González et al., 2008).

An important way of categorizing and stereotyping people is social class (Durante & Fiske, 2017; Lott, 2010). This is particularly important in societies with a high economic inequality, in which the attribution of competence or agency to high-class members is greater (Moreno-Bella et al., 2019). In fact, Cuddy et al. (2009) showed that rich people are perceived to be highly competent and low in warmth, whereas poor people are perceived to be low in both dimensions.

The association in the work context of status related stereotypes and certain racial groups (job-based race–status associations) and its effects has also been tested. Dupree et al. (2020) demonstrated that White Americans with higher implicit associations of White with high status (and Black with low status) showed more hierarchy-maintaining attitudes, beliefs, and preferences. This included higher prejudice, social dominance orientation, meritocracy, as well as policy preferences that support inequality.

Previous literature has studied with qualitative methods the influence of ethnicity and social class (among other categories) on discrimination experiences of Roma (e.g., Hellgren & Gabrielli, 2021; Kóczé, 2009). The present work will contribute to this field of research by analyzing with quantitative methods for the first time the effects of social class and ethnicity, separately as well as in intersection, on prejudice toward Roma. For this aim we will consider stereotypes (morality, sociability, and competence), emotions (positive and negative), and behavioral tendencies (facilitation, harm, and social distance). We will also measure job discrimination with explicit measures of intergroup relations.

Is it Racism or Classism?

One of the main difficulties that social scientists have is disentangling the effects of race/ethnicityFootnote 1 from the effects of class, especially considering that in many countries ethnic minorities are disproportionately found in the lower economic strata (Dupree et al., 2020). Thus, an important research question is to examine whether some instances of racism or ethnic prejudice can be merely vestiges of class-related processes, or consider that it could be an interactive effect between class and ethnicity.

Ethnicity represents, beyond phenotypic and genotypic associated characteristics, what people have learned about traditions and customs of their communities of origin. An ethnic group is associated with special experiences in language, music, history, literature, food, and celebrations considered to be similar to that of others with a shared background.

Social class—also called socioeconomic status in the literature (SES)—implies the material and social resources that an individual has, such as income, educational attainment, and occupational prestige (Kraus & Stephens, 2012). Wright (2009) defined the term class as the way of considering the interconnections between the individual attributes of people (e.g., education, cultural resources, or social connections) and the material conditions of life. When the two comes together, they create groupings called classes. For example, the so-called lower class is made up of people who do not have the educational or cultural resources to live above the poverty line (Wright, 2009). Importantly, Easterbrook et al. (2020) showed that people place a high importance on their SES identities, participants attached at least as much importance to their SES identities as they did to more traditional identities, such as ethnicity or gender.

Despite the fact that prejudice has been classically studied in relation to ethnic or gender group categories, people from the lower class is one of the groups that is more discriminated against. This has been called aporophobia by Cortina (2017), highlighting the idea that as it happens with homophobia or islamophobia, social class membership is crucial for understanding intergroup relations.

Social class is associated with access to the most basic resources, such as food, medical care, shelter, and education (Lott, 2010). As Lott (2012) argues:

Membership in a given social class (a) reliably predicts the degree to which one can obtain and benefit from a society’s economic and political resources, (b) is correlated with a wide array of life experiences, and (c) mediates and influences what a person is likely to learn, believe, anticipate, and seek after (p. 650).

Despite the fact that most research has focused on how singular social categories (e.g., race or social class) influence intergroup processes, some exceptions have considered the relevance of studying both categories or/and their potential intersection (e.g., Dupree et al., 2020; Mezzadra, 2021; Weeks & Lupfer, 2004). When trying to disentangle the joint effects of class and race categories on prejudice and discrimination, or its conjunction, there have been three main perspectives (Weeks & Lupfer, 2004).

First, it has been argued that it is race (or ethnicity) that has a stronger influence on prejudice (e.g., Pettigrew, 1980). Pettigrew (1980) showed that social class was increasing its influence on Blacks’ lives during the mid-1900s, but race continued to be an important factor above and beyond social class. This position is also supported by the ethnic-prominence hypothesis (Levin et al. 2002).

Second, other authors maintain that intergroup conflicts are mainly a class issue (e.g., Wilson, 1978). For example, Bayton et al. (1976) investigated the stereotypes associated with upper- and lower-class Blacks and Whites. They found that the stereotypes were a consequence of social class and not of the racial category. For example, the stereotypes typically associated with Blacks (e.g., lazy, unreliable, ignorant) were the most indicative of lower-class targets, for both black and white.

Thirdly, there are authors arguing that what is really happening is an interaction effect between the two categories (i.e., a Race/Ethnicity × Class). From this perspective, Kessler and Neighbors (1986) showed that the effects of race and social class on an individual’s psychological distress were not additive but interactive—race differences became more pronounced among lower-class individuals. Similarly, Weeks and Lupfer (2004) in two experimental studies showed that during spontaneous categorization of targets, race dominates categorization in stereotype-consistent race‐class associations (i.e., Black people from the lower class), but class influences categorization in stereotype‐inconsistent race‐class associations (i.e., middle class Blacks). Their results illustrate the importance of social class membership as a social category, independently but also in conjunction with race.

Overview of Studies

The general objective of the present research is to analyze the effect of social class and ethnicity on intergroup processes, such as prejudice and discrimination from the majority society toward Roma. We also aim to disentangle the predictive effect of class membership from the classic established predictive effect of ethnic membership and explore the intersection of both factors.

In Study 1, we began investigating our most basic question: Analyzing the effects of social class and ethnicity on prejudice and discrimination among a sample of undergraduate students. In Study 2, based on the obtained results in Study 1 showing a main effect of social class, we preregistered the hypotheses of the effects of social class in all dependent variables separately, conducted the study with a bigger sample of non-Roma general population in Spain, and explore the interaction effect between both independent variables. Finally, having confirmed that social class has a main effect on stereotypes, emotions and discrimination measures, and due to the fact that we found an interaction effect in Study 2, an integrative analysis with the two samples was conducted in order to confirm the interaction effect of ethnicity and social class among a sample with a bigger size. These studies analyze for the first time the effects of social class and ethnicity, separately as well as in intersection, on prejudice toward Roma considering the evaluation of stereotypes, emotions, and behavioral tendencies of discrimination.

Supplementary materials (Online Resources) and databases can be viewed on the OSF link: https://osf.io/jge6c/?view_only=34b1fa15d4d4471381cd46c7e5ddb0f2.

Study 1

Method

Participants and Design

A power analysis using G*Power 3.1 for an F test (ANOVA for fixed effects, special, main effects and interactions) showed that for obtaining a medium effect size (f = .25), with an alpha of 0.05 and power of .80, we would require 128 participants. We, therefore, established to collect a minimum of 130 and maximum of 150 observations after exclusions (1 participant was excluded for being Roma). One hundred and thirty-four (74% female) non-Roma university students voluntarily participated. They were residents in a city in the south of Spain with a high presence of Roma population. The age of the participants ranged from 18 to 53 years (M = 20.38; SD = 4.12). Participants were randomly assigned to one of four experimental conditions with a between subjects design of 2 (High vs. Low social class) × 2 (Ethnicity: Roma vs. Non-Roma), with 67 participants in each social class and 67 in each ethnic group condition.

Procedure and Instruments

The questionnaires, in paper and pencil format, were answered individually by the participants during approximately 20 min. Before starting the questionnaire, participants were informed about different aspects of the research: objectives, responsible people, voluntary participation, and the possibility of stopping at any time. Likewise, they were informed that the data obtained would be treated anonymously, confidentially, globally, and statistically. All of them signed an informed consent and received an extra grade in one of their courses. The study was approved by the Human Research Bioethics Committee of the researchers’ university.

Participants were randomly assigned to answer one of four types of questionnaires depending on the experimental condition. In all versions of the questionnaire a text was presented and subsequently the scales to measure the variables. Participants were told that information would be presented about a representative person of one profile (i.e., from a group of people with these characteristics) (These profiles can be seen in the supplementary materials at the Open Science Framework link), and then they would have to answer some questions related to it. This profile was supposedly extracted from results obtained in the CIS (National Centre of Sociological Research in Spain) surveys. Since Kraus and Stephen (2012) theoretically and empirically supported that educational attainment, income, and occupation represent the most widely used indices of objective social class, those three were used on the target profiles presented in the cover story in order to manipulate social class (High vs. Low). Ethnicity was directly mentioned and the surname of the person (representative of Roma vs. Non-Roma) was also presented in order to manipulate ethnicity. Then, participants answered the following measures of the variables:

Manipulation Check

In order to evaluate the target’s perceived social class, participants had to place the person described in the profile on one of the steps of a ladder (from 1 = The lowest social class; to 10 = The upper social class) (Kraus et al., 2009). We did not include a manipulation check of ethnicity because the ethnicity was explicitly mentioned in the profile presented to participants.

Stereotypes

A 9-item scale was used (López-Rodríguez et al., 2013), measuring three stereotypical dimensions (morality, sociability, and competence). Participants indicated to what extent they believed that people like those described in the profile were: honest, sincere and trustworthy (morality dimension), pleasant, friendly and warm (sociability dimension), intelligent, skilful and competent (competence dimension). The answer scale ranged from 1 (not at all) to 5 (very much). Average scores were calculated, ranging from 1 to 5, with high scores on each subscale indicating a more positive stereotype. The reliability was estimated using the Cronbach's alpha coefficient: .87 (for morality), .83 (sociability), and .83 (competence).

Intergroup Emotions

This variable was measured using a 24-item scale (Cuadrado et al., 2016). Participants were asked to indicate to what extent they felt the following emotions toward people like those described in the profile. Ten items measured positive emotions (e.g., admiration, understanding; α = .83), and 14 items measured negative emotions (e.g., disappointment, fear; α = .85). The answer scale ranged from 1 (not at all) to 5 (very much). Average scores were calculated, ranging from 1 to 5, with high scores on each subscale indicating higher intensity of emotions (positive or negative).

Behavioral Tendencies

This variable was measured with the 11-item Intergroup Behavioral Tendencies Scale (López-Rodríguez et al., 2016), based on the BIAS Map (Cuddy et al., 2008). Participants indicated to what extent they would be willing to carry out diverse actions toward people like those described in the profile. Five items measured facilitation tendencies (e.g., facilitating their promotion at work, α = .85), and six items measured harm tendencies (e.g., use derogatory nicknames to refer to them; α = .65). The answer scale ranged from 1 (not at all) to 5 (very much). Average scores were calculated, ranging from 1 to 5, with high scores on each subscale indicating a greater tendency toward facilitation or harm.

Preferred Social Distance

This variable was measured with a 5-item scale (adapted from Betancor et al., 2002). Participants indicated, “If you could choose, how far would you be willing to go in your relationship with people like those described in the profile?” Five items indicated different degrees of social distance or preferred intimacy in the relationship (from “starting a family” to “have them as neighbors”). The answer scale for each item ranged from 1 (not at all willing) to 5 (very willing). Item 5 (“I prefer not to have a relationship with one of those people features”) was eliminated from the analyses because the alpha went from .59 to .85, excluding it. Average scores were calculated (ranging from 1 to 5), with higher scores indicating a preference for maintaining closer relationships or less social distance with the target (α = .85).

Recommendation for a Job

This variable was measured with a single item: “To what extent would you recommend people like those described in the profile for a job?”. The answer scale ranged from 1 (I would not recommend them at all) to 5 (I would totally recommend them).

Socio-Demographic Variables

Participants indicated age, sex, place of birth, social class, belonging to the Roma group (Yes / No), mother's second surname, and date of birth.Footnote 2

Data Analyses

In addition to alpha coefficients and descriptive analyses of the studied variables, correlations and analyses of variance were carried out. Analyses of variance (ANOVAS) were conducted in order to confirm that the manipulations were effective. Main and interaction effects were conducted with an F test of variance using SPSS 20.

Results

The results of the manipulation check showed that participants who read the low-class profile attributed significantly lower social class (M = 2.51; SD = 1.02) to the person of the profile in comparison with participants assigned to the high-class condition (M = 8.15; SD = 1.45), F (1, 132) = 676.05, p < .001, η2p = .838.

Descriptive statistics and bivariate correlations between variables are presented in Online Resource 1 at the OSF link indicated previously. We found statistically significant correlations in the expected direction between all the dependent variables. Thus, the cognitive (stereotypical dimensions), affective (positive and negative emotions), and behavioral (behavioral tendencies, social distance and recommendation for a job) scales used to measure explicit intergroup attitudes, were significantly correlated in the expected direction.

As a result of the ANOVAS, the results of main effects of the manipulations of ethnicity and social class showed a significant effect of social class on almost all of the dependent variables (see Table 1). As expected, participants assigned to the experimental condition of low (vs. high) social class showed lower stereotypical perceptions of competence (Mlow = 3.10, SDlow = 0.95; Mhigh = 4.10, SDhigh = 0.73), less positive emotions (Mlow = 2.90, SDlow = 0.78; Mhigh = 3.52, SDhigh = 0.73), greater negative emotions (Mlow = 1.84, SDlow = 0.50; Mhigh = 1.49, SDhigh = 0.52), less willingness to help the target (Mlow = 3.04, SDlow = 0.80; Mhigh = 3.44, SDhigh = 0.86), greater tendency to harm the target (Mlow = 1.57, SDlow = 0.48; Mhigh = 1.30, SDhigh = 0.41), and they would prefer to maintain a greater social distance with the target (Mlow = 2.95, SDlow = 0.81; Mhigh = 4.02, SDhigh = 0.75) and would be less likely to recommend the target for a job (Mlow = 2.73, SDlow = 1.09; Mhigh = 4.27, SDhigh = 0.78). We did not find statistically significant differences between participants of the condition of low (vs. high) social class in the stereotypical perceptions of the target’s morality and sociability.

Table 1 Effect of social class and ethnicity in all dependent variables

A main effect of the manipulation of ethnicity was only significant on negative emotions, F (1,132) = 6.01, p = .016, η2p = .045, in which participants showed less negative emotions toward Roma (M = 1.56, SD = 0.49) than toward non-Roma targets (M = 1.77, SD = 0.56). No effect was found in any of the dimensions of stereotypes, in positive emotions, behavioral tendencies, social distance, or recommendation for a job.

The interaction between the two independent variables was not significant for any dependent variable (ps ≥ .300), with the exception of behavioral tendencies of facilitation, F (1, 132) = 4.20, p = .042, η2p = .032. Participants assigned to the low-class Roma condition show the least willingness to help (M = 2.98, SD = 0.79), whereas the participants assigned to the high–class Roma condition showed higher scores on facilitation (M = 3.68, SD = 0.90).Footnote 3

Discussion

This study aimed to analyze the effect of social class and ethnicity of Roma (the main derogated group in Spain) on prejudice and discrimination from the majority society toward this minority group.

As in some previous studies, results showed that social class had stronger effects than ethnic membership (Wilson, 1978). So, to perceive others as having low (vs. high) social class influences more than their ethnicity (Roma vs. Non-Roma) in how people stereotype them, feel about them, and behave toward them. In our study, low (vs. high) social class influenced target perceptions (less competent stereotype), emotions (eliciting less positive and more negative emotions), and behaviors toward the target (less willingness to help and more willingness to harm, greater preferred social distance, and less recommendation for a job). However, to perceive others as Roma vs. Non-Roma (ethnicity) only had an influence on more negative emotions. Finally, the interaction effect between class and ethnicity (e.g., Kessler & Neighbors, 1986; Weeks & Lupfer, 2004) was not supported in this study, except on one behavioral variable— tendency to help. However, a possible limitation of this study was that the sample size may not be sufficiently powered to detect an interaction effect.

According to the SCM (Fiske et al., 2002) and its extensions (Brambilla et al., 2012), we could expect that the low (vs. high) social class target would be perceived as stereotypically less competent but also less warmth (i.e., less moral and sociable). Predictions about competence were confirmed in our study, but we have found no difference in attributions of morality and sociability. Moreover, in line with the prediction that would be associated to the SCM in relation to social class, the present results support that low social class people are targets of more negative emotions (and less positive emotions) as well as more negative behavioral tendencies.

The SCM extensions (i.e., tri-dimensional stereotypical model) have never been applied to assess Roma targets in Spain; although, they have been applied to other devalued minority ethnocultural groups such as immigrants (e.g., López-Rodríguez et al., 2013). The results of those studies show variability in stereotypical dimensions (morality, sociability, and competence), depending on the immigrants’ background. In our study, however, ethnicity (Roma vs. Non-Roma) had no effect on any of the stereotypical dimensions.

In order to deepen understanding regarding the predictive power of social class membership and to analyze whether this main effect was replicated, we carried out a new preregistered experimental study with a greater sample size and composed of the general population and not just of undergraduate students.

Study 2

The main hypotheses, based on the results of the previous study, were preregistered in the Open Science Framework, OSF (See https://osf.io/jge6c/?view_only=34b1fa15d4d4471381cd46c7e5ddb0f2), and were the following:

Participants assigned to the low-class experimental condition will show lower scores on the competence dimension of stereotypes toward the target group than those assigned to the high-class experimental condition (Hypothesis 1).

Participants assigned to the low-class experimental condition will show more negative emotions toward the target group (H2) and less positive emotions toward the target group (H3) than those assigned to the high-class experimental condition.

Participants assigned to the low-class experimental condition will show more negative behavioral tendencies (lower facilitation and higher harm) toward the target group (H4), will recommend target group members less often for a job (H5), and will show a greater social distance toward the target group (H6) than those assigned to the high-class experimental condition.

In addition, we explored the interaction effects of Ethnicity × Class in all dependent variables with a larger sample size.

Method

The sample size was determined by the criterion of collecting at least 256 participants (64 participants per condition, estimating a power of .80 and an effect size of d = 0.50).

The sample was collected in a simple random way through online questionnaires on social networks (Twitter, Facebook, and Instagram) due to the state of confinement in Spain in 2020–2021 derived from the pandemic. In this way, people who agreed to participate in the study were only aware that it was a study about social perceptions. Participants who did not pass the pre-established criteria—i.e., those who did not exceed more than 50% of the survey or who were Roma (only 1 participant was Roma) were excluded from the analyses. The final sample was composed of 282 participants from the general population (60.2% women), with ages between 18 and 72 years (M = 30.5, SD = 12.4). The level of studies of the participants consisted of 5.1% with primary education, 10.7% with secondary education, 16.7% with a baccalaureate level, 50% with a university degree or bachelor's degree, 15.8% with a master's level degree, and 1.7% with a doctorate or higher.

In relation to the employment situation of the participants, 17.1% were self-employed, 23.1% worked with a fixed contract, 14.1% worked with a temporary contract, 33.8% were students, 0.4% were retired, and 11.5% were unemployed.

Participants were randomly assigned to one of the four experimental conditions with a design between subjects 2 (High vs. Low social class) × 2 (Ethnicity: Roma vs. Non-Roma). In this study, we made sure to have more than 64 participants in each of the four conditions before conducting the main analyses.

Procedure, instruments,Footnote 4 and data analysis were the same as in Study 1.

The reliability coefficients (Cronbach's alpha) and descriptive statistics of the measures are presented in Online Resource 2 at the OSF link.

Results

Firstly, the manipulation check showed that participants assigned to the low-class condition significantly attributed a lower social class to such a profile (M = 3.07; SD = 1.67) than participants assigned to the high social class condition (M = 7.17; SD = 2.03), F (1, 281) = 345.52, p <.001, η2p = .552.

Bivariate Correlations

The results of correlational analyses between dependent variables are presented in Online Resource 3 at the OSF link. We found statistically significant correlations in the expected direction between all the dependent variables.

Effects of Social Class, Ethnicity and Interaction of Class × Ethnicity

As preregistered, we conducted ANOVA analyses and the results showed a significant main effect of social class on almost all of the dependent variables (see Table 2), replicating the effects of class obtained in Study 1. As expected, participants assigned to the experimental condition of low (vs. high) social class showed lower stereotypical perceptions of competence (H1), more negative emotions (H2), and less positive emotions (H3); they would recommend the target less often for a job (H5) and would prefer to maintain a greater social distance with the target (H6). We did not find statistically significant differences between participants of the condition of low (vs. high) social class in the stereotypical perception of the target's morality and sociability and in behavioral tendencies of facilitation and harm.

Additionally, we found a significant interaction effect of Class × Ethnicity in all the dependent variables (see Table 2) which we consider an important contribution to the field of prejudice.Footnote 5

Table 2 Effects of social class, ethnicity, and interaction of class × ethnicity (Study 2)

In the post hoc pairwise comparisons (Bonferroni), according to the ethnicity condition, it can be seen as in the case of Roma targets, class differences have a greater weight than that of non-Roma targets on most dependent variables (see Table 3 and Fig. 1), with the exceptions of stereotypes of morality and sociability as well as the behavioral tendency of facilitation in which, although the direction is opposed for Roma and non-Roma, the weight is higher for non-Roma targets.

Table 3 Effects of social class for Roma and non-Roma targets in all dependent variables
Fig. 1
figure 1

Mean scores on dependent variables by experimental condition (Study 2)

Discussion

The hypotheses based on the main effect of social class were mainly confirmed, since there were significant differences in attribution of competence stereotypes, positive and negative emotions, social distance, and recommendation for a job, thus, confirming Hypotheses 1, 2, 3, 5, and 6. However, Hypothesis 4 was not corroborated, which predicted effects on facilitation and harm behaviors. These results replicated the importance of social class when evaluating Roma.

According to the SCM (Fiske et al., 2002) and BIAS Map (Cuddy et al., 2008), predictions about competence and emotions were confirmed in our study (a low-class target was perceived as less competent, and elicited more negative emotions and less positive emotions), but we have found no difference in attributions of morality and sociability or in behavioral tendencies of facilitation and harm.

Importantly, in this study we found an interaction effect between the ethnic category and the social class of the target, supporting the literature that defend the importance of the interaction between class and ethnicity (Kessler & Neighbors, 1986; Weeks & Lupfer, 2004). These results suggest that both variables may act in conjunction in an interactive process—and not in an isolated or additive form.

In short, results showed that the effects of social class were stronger for Roma targets than for non-Roma targets. Put differently, belonging to a low social class had more costs for the Roma than for the ethnic majority group (non-Roma). This pattern showed up in the most blatant measures of prejudice and discrimination: more negative emotions (and less positive), more harm tendencies, more social distance, and less recommendations for a job. As observed in Table 3 and Fig. 1, whereas Roma of lower class are negatively evaluated and more discriminated than any other group, Roma of higher class are positively evaluated. This happens more for Roma than for non-Roma targets of higher class. This pattern was found for the stereotype of competence, but it was not for morality and sociability; in this case, social class was more important for non-Roma than for Roma people. The same happened for behavioral tendencies of facilitation.

To obtain a greater statistical power for testing the interaction effect of Class × Ethnicity, we analyzed both samples together using a pooled data analysis (i.e., an integrative data analysis; Curran & Hussong, 2009). So, we examined whether these effects are replicated.

Integrative Data Analyses

Whereas the results of Studies 1 and 2 are highly symmetrical, replicating a main effect of social class in most of dependent variables, there was some discrepancy between the results on the interaction effect: it was not significant in Study 1, whereas it was significant for all dependent variables in Study 2 as well as some differences on the main effect of ethnicity. For this reason, to test the interaction of class and ethnicity and to clarify the potential discrepancies considering all the data, we conducted an integrative data analysis (Curran & Hussong, 2009). This procedure allowed us to increase the statistical power for conducting analyses of the 2 × 2 interaction as well as sample heterogeneity. We pooled the samples from the two studies into a single multivariate analysis of variance to confirm main and interaction effects. The total sample included 416 participants (N1 = 134; N2 = 282).

The multivariate analyses of variance with ethnicity condition (Roma vs. Non-Roma), class condition (Low vs. High), and Study (1 vs. 2) as fixed factors showed that the triple interaction (Ethnicity × Class × Study) was not significant for any dependent variable (ps between .070 and .860). Importantly, results corroborated the interaction between ethnicity and class for most of variables (see Table 4), since we found a significant interaction effect in almost all dependent variables, except for the competence stereotype and behavioral tendencies of facilitation, which were both predicted only by social class. To confirm a significant interaction effect in almost all dependent variables in the pooled data supports the idea that the lack of significance in Study 1 could be due to the smaller sample size.

Table 4 Between-subject effects of social class, ethnicity, and interaction of class × ethnicity (pooled data)

In the post hoc pairwise comparisons (Bonferroni) according to the ethnicity condition, it can be seen as in the case of Roma targets, class differences have a greater weight than that of non-Roma targets on most dependent variables (see Online Resource 4 and Fig. 2).

Fig. 2
figure 2

Mean scores of dependent variables in pooled data (standardized scores)

As in Study 2, the effect of class is higher for Roma targets than for non-Roma targets, although mean differences are significant for both ethnic groups in many variables, the weight of class is more acute for Roma than for non-Roma. Participants assigned to the Roma high-class condition showed the most positive responses (attribution of positive stereotypes, more positive emotions, higher facilitation, lower harm, lower social distance, and more recommendations for a job), and the participants assigned to the low-class Roma showed the lowest attribution of morality, less positive emotions toward people like the one presented in the profile, the highest score of behavioral tendencies of harm, the highest social distance, and the lowest level of recommendation for a job (see Fig. 2).

General Discussion and Conclusions

The main objective of this work was to analyze the effect of social class and ethnicity of Roma on prejudice and discrimination from the majority society toward this minority group.

One of the main difficulties that social scientists have is disentangling race/ethnicity from class, especially considering that in many countries, ethnic minorities are found disproportionately in the lower economic strata. Using the definition and indicators of social class provided by Kraus and Stephens (2012), the present work tries to answer research questions that need to be approached: to what extent are the examples that have been considered racism or ethnic prejudice cited also based on class-related processes?

With this aim, two experimental studies with different samples were carried out in which ethnicity (Roma vs. Non-Roma) and social class (Low vs. High) were manipulated. The Roma minority was chosen as the target group because it is a traditionally discriminated against minority in Europe and the most discriminated against in Spain.

Three main perspectives have been considered in the literature to answer the question of disentangling ethnic/racial prejudice from classism (Weeks & Lupfer, 2004): (1) some authors defend that it is race (or ethnicity) that mainly has influence in prejudice (e.g., Pettigrew, 1980); (2) some authors defend that social class has higher explicative capacity, contending that intergroup conflicts are many times due to a class issue more than to ethnic or racial membership (e.g., Wilson, 1978); (3) some authors defend that it is really an interaction between the two categories (race/ethnicity × social class) that matters (e.g., Kessler & Neighbors, 1986; Weeks & Lupfer, 2004).

The first perspective is not supported by our results. Although we found consistent effects of social class on Study 1, and this was then replicated in a preregistered experiment (Study 2), we considered that the interaction between class and ethnicity can better explain our pattern of results: Study 2 found a significant interaction effect for most variables, and this effect was replicated when using an integrative data analysis (Curran & Hussong, 2009).

The relevance of the interaction effects on prejudice lead us to discuss about the intersectional hypotheses of prejudice and discrimination as well as to introduce the concept of intersectionality (for results supporting intersectionality in Social Psychology see Kessler & Neighbors, 1986; Mattan et al., 2019; Moore-Berg & Karpinski, 2019; Weeks & Lupfer, 2004). Intersectionality is a theoretical approach that simultaneously considers multiple categories of identity, such as gender, race, social class, sexual orientation, etc. (Cole, 2009). This approach is originally based on critical race and feminist theories (Crenshaw, 1989), criticizing the tendency to consider some social categories, such as those based on race, to be homogeneous. In fact, members vary substantially on other relevant dimensions, such as social class and gender (see Moore-Berg & Karpinski, 2019, for a review of the effects of intersectionality in the psychological literature). Mattan et al. (2019) demonstrated a similar finding in basic experiments in implicit evaluative bias where participants were most sensitive to target SES, but also showed interactive effects by target’s race.

In short, our results show that perceiving a person (or group) with a low or high social class (i.e., indicated by socioeconomic level of neighborhood, education level, occupation, and income level), leads people to attribute them certain stereotypes, feel certain emotions, or be willing to carry out certain behaviors. This happens even more for ethnic minorities.

As such, the effects of social class were more important for Roma individuals than for non-Roma in most variables. Specifically, low-class Roma (vs. high-class Roma) were perceived as less moral and sociable; people felt less positive and more negative emotions toward them; and they elicited greater behavioral tendencies of harm, greater social distance, and were less often recommended for a job. All these social class effects did not occur, or were less strong, when comparing low- and high-class non-Roma individuals.

All in all, these results suggest that it is important to simultaneously consider multiple identity categories (Cole, 2009). Thus, if we want to examine the effects of social class on how people are evaluated, it is also important to consider their other group memberships. Previous studies have found that, when examining intergroup relations in the USA, “There is more to being a lower class black than simply being black and lower class” (Weeks & Lupfer, 2004, p. 974), or in words of Crenshaw (1989) referring to black women:

Imagine a basement which contains all people who are disadvantaged on the basis of race, sex, class, sexual preference... These people are stacked-feet standing on shoulders-with those on the bottom being disadvantaged by the full array of factors, up to the very top, where the heads of all those disadvantaged by a singular factor brush up against the ceiling. Their ceiling is actually the floor above which only those who are not disadvantaged in any way reside. In efforts to correct some aspects of domination, those above the ceiling admit from the basement only those who can say that “but for” the ceiling, they too would be in the upper room. A hatch is developed through which those placed immediately below can crawl… Those who are multiply-burdened are generally left below unless they can somehow pull themselves into the groups that are permitted to squeeze through the hatch. (pp. 151–152)

The present studies show that a similar pattern occurs when we examine different components or expressions of prejudice and discrimination toward lower-class Roma people in Spain.

However, the lingering questions are: what is the nature of this Ethnicity × Class interaction and what mechanisms may be driving it? Why might high-class Roma people be evaluated so positively and low-class Roma people be evaluated so negatively? One possible explanation for positive responses toward high-class Roma targets is that a process of subtyping is taking place. The motivation to maintain presently held beliefs about ethnic groups often leads people to search for a justification that does not generate cognitive dissonance. One means of maintaining these beliefs is to subtype by classifying the stereotype-disconfirming outgroup member as an “unusual case” or subtype of the larger group (Richards & Hewstone, 2001). Participants might consider Roma people pertaining to a high class as not representative of the outgroup, thus, not applying the negative representation and affect that endorse toward the rest of the outgroup.

Another explanation, and in some way connected to the previous one, could be related to participants’ meritocracy beliefs. Meritocracy beliefs have a deep impact on the way people think about and act toward low-status groups (Madeira et al., 2019). Since Roma people occupying high privileged positions are perceived as uncommon, meritocracy beliefs could simultaneously explain the positive evaluations and behaviors toward high-class Roma—they individually achieved a good socioeconomic position regardless the starting point of their group—and the negative evaluations and discrimination toward low-class Roma targets—they are in that situation because they don’t make enough effort for changing their reality—. Meritocracy is increasingly associated with intolerance and dislike of members of low-status groups (Madeira et al., 2019), overrepresented by ethnic and racial minorities (Dupree et al., 2020).

This work presents some limitations. First of all, the sample of Study 1 was small, and we may not have had enough power to detect the interaction effect. Despite the fact that the pooled analyses support the intersectionality perspective, it would be convenient to carry out new research with larger samples in relation to other minority groups (e.g., immigrants or indigenous people).

Secondly, these results may also be related with what Cole (2009) refers as an intersectional conceptualization of multiple categories on psychosocial research, future studies could look at presenting an underrepresented group perspective defined by multiple social identities and to situate the intersection of ethnic and class identities in different contexts, both through the priming manipulation and the cross-cultural approach. However, we are conscious that our methods did not allow us to conduct intersectional research and that observing a significant interaction effect of different social categories is a necessary, yet not sufficient, condition for intersectionality to occur (see Bowleg & Bauer, 2016). Future designs could incorporate other types of group-level population, environmental, and policy variables to advance understanding about social–structural power relations and their differential effects at varying intersections as well as use qualitative methods in addition to the quantitative methodology. However, these results are an essential starting point for putting attention to the intersectionality in psychological research for understanding prejudice toward minorities because compared with many disciplines, intersectionality remains relatively inchoate within psychology (Bowleg & Bauer, 2016).

Despite these limitations, we consider that this work makes important contributions to the field of prejudice and majority–minority intergroup relations. The old debate about Classism vs. Racism has tipped the research fundamentally in favor of racial or ethnic membership as determinants of prejudiced attitudes and discriminatory behaviors. This has led to a neglect of the importance of social class as a social category with equal or even greater effects on intergroup prejudice. This also leads us to forget that ethnic or racial minorities are also often lower class.

Results obtained in our work support the importance of considering social class discourses in order to find resistance strategies and effective prejudice reduction interventions that take into consideration more than one membership at the same time. As such, it will important that interventions aimed to increase social class diversity also take into account race/ethnic membership. It will not be the same to participate in these programs when people are a member of an ethnic minority or not.

In conclusion, this paper presents two studies that consistently found that class is an important variable that influence how Roma people are evaluated in Spain, especially in interaction with ethnicity. Considering the intersection of different categories of exclusion is useful for policy recommendations. Understanding how these groups are evaluated by the majority group is a first step for developing more inclusive and egalitarian programs and policies that could improve intergroup relations.